Video2NAND – Abusing video codecs for great computational power
2 days ago
- The article explores building logic circuits using VP8 video codec prediction mechanisms instead of transistors.
- Video codecs encode video frames, with key-frames being independent and inter-frames referencing previous ones.
- In this setup, white blocks represent True (255) and black blocks represent False (0).
- Wires are modeled using H_PRED (horizontal prediction) and V_PRED (vertical prediction) modes.
- The TM_PRED mode is used to construct NOT and AND gates, forming a functionally complete set.
- A NOT gate is implemented as 255 - input by setting top row to 255 and left column to 0.
- An AND gate is made by setting top_left to 255 and using top row and left column as inputs.
- These gadgets can be combined to create more complex gates like NAND.
- Open questions include optimizing gadgets, integrating with synthesis tools, and exploring sequential logic with inter-frames.